High-Performance Data Analytics Techniques for Power Markets Simulation

Efficient energy trading relies on high-fidelity price prediction systems with short response times that enable producers, consumers, and traders to make informed decisions in real time. Electricity networks are typically modeled using linear programming (LP), with the uncertainty in supply and demand being represented by hundreds of thousands of different input parameters. Processing large sets of probabilistic scenarios in near real-time is imperative to effective trading. High-throughput scheduling techniques of the individual LPs provide a promising and often overlooked opportunity to reduce processing time. This work analyzes high-throughput computational techniques in the context of European electricity markets. An LP model of the electricity price that considers different load scenarios, marginal cost functions, and weather patterns is introduced and analyzed from the perspective of computational resource requirements for massively parallel simulations. We present numerical experiments that consider multiple aspects of the computational pipeline from the choice of the LP solution algorithm and its impact on memory efficiency, to the scheduling techniques that allocate individual jobs to computational resources. Results show that an optimized high-throughput computation strategy can reduce response times by 63% over a naive strategy in a real-world trading scenario.